The creative ghost in the algorithm: AI-generated photo-based images

Abstract AI-generated photo-based images are produced with generative AI systems that take user-uploaded photographs as visual prompts (optionally combined with verbal prompts). The paper maps current debates about AI and creative autonomy, contrasting autonomy-optimist views that treat AI as a creator or co-creator with autonomy-pessimist views that regard it as an advanced tool, while also noting the possibility of a non-binary, spectrum conception of autonomy. It argues that these images inherit photographic indexicality in their constituent parts (understood functionally, not in terms of material identity) while lacking indexicality as wholes, because no single scene ever served as the indexical cause of the overall image. They are therefore quasi-indexical compound images, ontologically akin to multiple exposures, photomontages, and certain long-exposure practices rather than to fully indexical photographs or non-photographic images. Situating the practice within other media where creative autonomy is partial or shared, the paper claims that users retain intentional control over conception, prompting, iteration, and selection, while ceding some production details to partially predictable algorithmic “glitch” processes. On this basis, it concludes that AI-generated photo-based images are quasi-indexical images produced by partially predictable and controllable glitch processes in which human creative autonomy is partial. It maintains that, insofar as other quasi-indexical images are recognised as photographic practices, AI-generated photo-based images should likewise be recognised as a photographic practice within the medium of photography. The paper closes by locating human creativity across pre-engagement, engagement, and post-engagement phases of image production, and by outlining directions for interdisciplinary research on algorithmic and interactive human-algorithmic creativity under conditions of meaningful human control.